2 * Copyright 2000, International Business Machines Corporation and others.
5 * This software has been released under the terms of the IBM Public
6 * License. For details, see the LICENSE file in the top-level source
7 * directory or online at http://www.openafs.org/dl/license10.html
9 * Portions Copyright (c) 2006 Sine Nomine Associates
13 * NEW callback package callback.c (replaces vicecb.c)
14 * Updated call back routines, NOW with:
16 * Faster DeleteVenus (Now called DeleteAllCallBacks)
17 * Call back breaking for volumes
18 * Adaptive timeouts on call backs
19 * Architected for Multi RPC
20 * No locks (currently implicit vnode locks--these will go, to)
21 * Delayed call back when rpc connection down.
22 * Bulk break of delayed call backs when rpc connection
24 * Strict limit on number of call backs.
26 * InitCallBack(nblocks)
27 * Initialize: nblocks is max number # of file entries + # of callback entries
28 * nblocks must be < 65536
29 * Space used is nblocks*16 bytes
30 * Note that space will be reclaimed by breaking callbacks of old hosts
32 * time = AddCallBack(host, fid)
34 * Returns the expiration time at the workstation.
36 * BreakCallBack(host, fid)
37 * Break all call backs for fid, except for the specified host.
40 * BreakVolumeCallBacksLater(volume)
41 * Break all call backs on volume, using single call to each host
42 * Delete all the call backs.
44 * DeleteCallBack(host,fid)
45 * Delete (do not break) single call back for fid.
47 * DeleteFileCallBacks(fid)
48 * Delete (do not break) all call backs for fid.
50 * DeleteAllCallBacks(host)
51 * Delete (do not break) all call backs for host.
53 * CleanupTimedOutCallBacks()
54 * Delete all timed out call back entries
55 * Must be called periodically by file server.
57 * BreakDelayedCallBacks(host)
58 * Break all delayed call backs for host.
59 * Returns 1: one or more failed, 0: success.
61 * PrintCallBackStats()
62 * Print statistics about call backs to stdout.
64 * DumpCallBacks() ---wishful thinking---
65 * Dump call back state to /tmp/callback.state.
66 * This is separately interpretable by the program pcb.
68 * Notes: In general, if a call back to a host doesn't get through,
69 * then HostDown, supplied elsewhere, is called. BreakDelayedCallBacks,
70 * however, does not call HostDown, but instead returns an indication of
71 * success if all delayed call backs were finally broken.
73 * BreakDelayedCallBacks MUST be called at the first sign of activity
74 * from the host after HostDown has been called (or a previous
75 * BreakDelayedCallBacks failed). The BreakDelayedCallBacks must be
76 * allowed to complete before any requests from that host are handled.
77 * If BreakDelayedCallBacks fails, then the host should remain
78 * down (and the request should be failed).
80 * CleanupCallBacks MUST be called periodically by the file server for
81 * this package to work correctly. Every 5 minutes is suggested.
84 #include <afsconfig.h>
85 #include <afs/param.h>
90 #ifdef HAVE_SYS_FILE_H
96 #include <afs/nfs.h> /* yuck. This is an abomination. */
98 #include <rx/rx_queue.h>
99 #include <afs/afscbint.h>
100 #include <afs/afsutil.h>
101 #include <afs/ihandle.h>
102 #include <afs/partition.h>
103 #include <afs/vnode.h>
104 #include <afs/volume.h>
105 #include "viced_prototypes.h"
108 #include <afs/ptclient.h> /* need definition of prlist for host.h */
110 #include "callback.h"
111 #ifdef AFS_DEMAND_ATTACH_FS
112 #include "serialize_state.h"
113 #endif /* AFS_DEMAND_ATTACH_FS */
116 extern afsUUID FS_HostUUID;
117 extern int hostCount;
119 #ifndef INTERPRET_DUMP
120 static int ShowProblems = 1;
123 struct cbcounters cbstuff;
125 static struct FileEntry * FE = NULL; /* don't use FE[0] */
126 static struct CallBack * CB = NULL; /* don't use CB[0] */
128 static struct CallBack * CBfree = NULL;
129 static struct FileEntry * FEfree = NULL;
132 /* Time to live for call backs depends upon number of users of the file.
133 * TimeOuts is indexed by this number/8 (using TimeOut macro). Times
134 * in this table are for the workstation; server timeouts, add
137 static int TimeOuts[] = {
138 /* Note: don't make the first entry larger than 4 hours (see above) */
139 4 * 60 * 60, /* 0-7 users */
140 1 * 60 * 60, /* 8-15 users */
141 30 * 60, /* 16-23 users */
142 15 * 60, /* 24-31 users */
143 15 * 60, /* 32-39 users */
144 10 * 60, /* 40-47 users */
145 10 * 60, /* 48-55 users */
146 10 * 60, /* 56-63 users */
147 }; /* Anything more: MinTimeOut */
149 /* minimum time given for a call back */
150 #ifndef INTERPRET_DUMP
151 static int MinTimeOut = (7 * 60);
154 /* Heads of CB queues; a timeout index is 1+index into this array */
155 static afs_uint32 timeout[CB_NUM_TIMEOUT_QUEUES];
157 static afs_int32 tfirst; /* cbtime of oldest unexpired call back time queue */
160 /* 16 byte object get/free routines */
165 /* Prototypes for static routines */
166 static struct FileEntry *FindFE(AFSFid * fid);
168 #ifndef INTERPRET_DUMP
169 static struct CallBack *iGetCB(int *nused);
170 static int iFreeCB(struct CallBack *cb, int *nused);
171 static struct FileEntry *iGetFE(int *nused);
172 static int iFreeFE(struct FileEntry *fe, int *nused);
173 static int TAdd(struct CallBack *cb, afs_uint32 * thead);
174 static int TDel(struct CallBack *cb);
175 static int HAdd(struct CallBack *cb, struct host *host);
176 static int HDel(struct CallBack *cb);
177 static int CDel(struct CallBack *cb, int deletefe);
178 static int CDelPtr(struct FileEntry *fe, afs_uint32 * cbp,
180 static afs_uint32 *FindCBPtr(struct FileEntry *fe, struct host *host);
181 static int FDel(struct FileEntry *fe);
182 static int AddCallBack1_r(struct host *host, AFSFid * fid, afs_uint32 * thead,
183 int type, int locked);
184 static void MultiBreakCallBack_r(struct cbstruct cba[], int ncbas,
185 struct AFSCBFids *afidp);
186 static int MultiBreakVolumeCallBack_r(struct host *host,
187 struct VCBParams *parms, int deletefe);
188 static int MultiBreakVolumeLaterCallBack(struct host *host, void *rock);
189 static int GetSomeSpace_r(struct host *hostp, int locked);
190 static int ClearHostCallbacks_r(struct host *hp, int locked);
191 static int DumpCallBackState_r(void);
194 #define GetCB() ((struct CallBack *)iGetCB(&cbstuff.nCBs))
195 #define GetFE() ((struct FileEntry *)iGetFE(&cbstuff.nFEs))
196 #define FreeCB(cb) iFreeCB((struct CallBack *)cb, &cbstuff.nCBs)
197 #define FreeFE(fe) iFreeFE((struct FileEntry *)fe, &cbstuff.nFEs)
200 /* Other protos - move out sometime */
201 void PrintCB(struct CallBack *cb, afs_uint32 now);
203 static afs_uint32 HashTable[FEHASH_SIZE]; /* File entry hash table */
205 static struct FileEntry *
210 struct FileEntry *fe;
212 hash = FEHash(fid->Volume, fid->Unique);
213 for (fei = HashTable[hash]; fei; fei = fe->fnext) {
215 if (fe->volid == fid->Volume && fe->unique == fid->Unique
216 && fe->vnode == fid->Vnode && (fe->status & FE_LATER) != FE_LATER)
222 #ifndef INTERPRET_DUMP
224 static struct CallBack *
227 struct CallBack *ret;
229 if ((ret = CBfree)) {
230 CBfree = (struct CallBack *)(((struct object *)ret)->next);
237 iFreeCB(struct CallBack *cb, int *nused)
239 ((struct object *)cb)->next = (struct object *)CBfree;
245 static struct FileEntry *
248 struct FileEntry *ret;
250 if ((ret = FEfree)) {
251 FEfree = (struct FileEntry *)(((struct object *)ret)->next);
258 iFreeFE(struct FileEntry *fe, int *nused)
260 ((struct object *)fe)->next = (struct object *)FEfree;
266 /* Add cb to end of specified timeout list */
268 TAdd(struct CallBack *cb, afs_uint32 * thead)
271 (*thead) = cb->tnext = cb->tprev = cbtoi(cb);
273 struct CallBack *thp = itocb(*thead);
275 cb->tprev = thp->tprev;
279 thp->tprev = (itocb(thp->tprev)->tnext = cbtoi(cb));
281 thp->tprev = cbtoi(cb);
284 cb->thead = ttoi(thead);
288 /* Delete call back entry from timeout list */
290 TDel(struct CallBack *cb)
292 afs_uint32 *thead = itot(cb->thead);
294 if (*thead == cbtoi(cb))
295 *thead = (*thead == cb->tnext ? 0 : cb->tnext);
296 if (itocb(cb->tprev))
297 itocb(cb->tprev)->tnext = cb->tnext;
298 if (itocb(cb->tnext))
299 itocb(cb->tnext)->tprev = cb->tprev;
303 /* Add cb to end of specified host list */
305 HAdd(struct CallBack *cb, struct host *host)
307 cb->hhead = h_htoi(host);
309 host->cblist = cb->hnext = cb->hprev = cbtoi(cb);
311 struct CallBack *fcb = itocb(host->cblist);
313 cb->hprev = fcb->hprev;
314 cb->hnext = cbtoi(fcb);
315 fcb->hprev = (itocb(fcb->hprev)->hnext = cbtoi(cb));
320 /* Delete call back entry from host list */
322 HDel(struct CallBack *cb)
324 afs_uint32 *hhead = &h_itoh(cb->hhead)->cblist;
326 if (*hhead == cbtoi(cb))
327 *hhead = (*hhead == cb->hnext ? 0 : cb->hnext);
328 itocb(cb->hprev)->hnext = cb->hnext;
329 itocb(cb->hnext)->hprev = cb->hprev;
333 /* Delete call back entry from fid's chain of cb's */
334 /* N.B. This one also deletes the CB, and also possibly parent FE, so
335 * make sure that it is not on any other list before calling this
338 CDel(struct CallBack *cb, int deletefe)
341 struct FileEntry *fe = itofe(cb->fhead);
345 for (safety = 0, cbp = &fe->firstcb; *cbp && *cbp != cbi;
346 cbp = &itocb(*cbp)->cnext, safety++) {
347 if (safety > cbstuff.nblks + 10) {
348 ViceLogThenPanic(0, ("CDel: Internal Error -- shutting down: "
349 "wanted %d from %d, now at %d\n",
350 cbi, fe->firstcb, *cbp));
351 DumpCallBackState_r();
352 ShutDownAndCore(PANIC);
355 CDelPtr(fe, cbp, deletefe);
359 /* Same as CDel, but pointer to parent pointer to CB entry is passed,
360 * as well as file entry */
361 /* N.B. This one also deletes the CB, and also possibly parent FE, so
362 * make sure that it is not on any other list before calling this
364 static int Ccdelpt = 0, CcdelB = 0;
367 CDelPtr(struct FileEntry *fe, afs_uint32 * cbp,
380 if ((--fe->ncbs == 0) && deletefe)
386 FindCBPtr(struct FileEntry *fe, struct host *host)
388 afs_uint32 hostindex = h_htoi(host);
393 for (safety = 0, cbp = &fe->firstcb; *cbp; cbp = &cb->cnext, safety++) {
394 if (safety > cbstuff.nblks) {
395 ViceLog(0, ("FindCBPtr: Internal Error -- shutting down.\n"));
396 DumpCallBackState_r();
397 ShutDownAndCore(PANIC);
400 if (cb->hhead == hostindex)
406 /* Delete file entry from hash table */
408 FDel(struct FileEntry *fe)
411 afs_uint32 *p = &HashTable[FEHash(fe->volid, fe->unique)];
413 while (*p && *p != fei)
414 p = &itofe(*p)->fnext;
421 /* initialize the callback package */
423 InitCallBack(int nblks)
426 tfirst = CBtime(time(NULL));
427 /* N.B. The "-1", below, is because
428 * FE[0] and CB[0] are not used--and not allocated */
429 FE = ((struct FileEntry *)(calloc(nblks, sizeof(struct FileEntry))));
431 ViceLogThenPanic(0, ("Failed malloc in InitCallBack\n"));
433 FE--; /* FE[0] is supposed to point to junk */
434 cbstuff.nFEs = nblks;
436 FreeFE(&FE[cbstuff.nFEs]); /* This is correct */
437 CB = ((struct CallBack *)(calloc(nblks, sizeof(struct CallBack))));
439 ViceLogThenPanic(0, ("Failed malloc in InitCallBack\n"));
441 CB--; /* CB[0] is supposed to point to junk */
442 cbstuff.nCBs = nblks;
444 FreeCB(&CB[cbstuff.nCBs]); /* This is correct */
445 cbstuff.nblks = nblks;
446 cbstuff.nbreakers = 0;
452 XCallBackBulk_r(struct host * ahost, struct AFSFid * fids, afs_int32 nfids)
454 struct AFSCallBack tcbs[AFSCBMAX];
460 struct rx_connection *cb_conn = NULL;
462 rx_SetConnDeadTime(ahost->callback_rxcon, 4);
463 rx_SetConnHardDeadTime(ahost->callback_rxcon, AFS_HARDDEADTIME);
469 for (i = 0; i < nfids && i < AFSCBMAX; i++) {
470 tcbs[i].CallBackVersion = CALLBACK_VERSION;
471 tcbs[i].ExpirationTime = 0;
472 tcbs[i].CallBackType = CB_DROPPED;
474 tf.AFSCBFids_len = i;
475 tf.AFSCBFids_val = &(fids[j]);
479 tc.AFSCBs_val = tcbs;
481 cb_conn = ahost->callback_rxcon;
482 rx_GetConnection(cb_conn);
484 code |= RXAFSCB_CallBack(cb_conn, &tf, &tc);
485 rx_PutConnection(cb_conn);
493 /* the locked flag tells us if the host entry has already been locked
494 * by our parent. I don't think anybody actually calls us with the
495 * host locked, but here's how to make that work: GetSomeSpace has to
496 * change so that it doesn't attempt to lock any hosts < "host". That
497 * means that it might be unable to free any objects, so it has to
498 * return an exit status. If it fails, then AddCallBack1 might fail,
499 * as well. If so, the host->ResetDone should probably be set to 0,
500 * and we probably don't want to return a callback promise to the
501 * cache manager, either. */
503 AddCallBack1(struct host *host, AFSFid * fid, afs_uint32 * thead, int type,
511 if (!(host->hostFlags & HOSTDELETED))
512 retVal = AddCallBack1_r(host, fid, thead, type, 1);
522 AddCallBack1_r(struct host *host, AFSFid * fid, afs_uint32 * thead, int type,
525 struct FileEntry *fe;
526 struct CallBack *cb = 0, *lastcb = 0;
527 struct FileEntry *newfe = 0;
528 afs_uint32 time_out = 0;
529 afs_uint32 *Thead = thead;
530 struct CallBack *newcb = 0;
533 cbstuff.AddCallBacks++;
537 /* allocate these guys first, since we can't call the allocator with
538 * the host structure locked -- or we might deadlock. However, we have
539 * to avoid races with FindFE... */
540 while (!(newcb = GetCB())) {
541 GetSomeSpace_r(host, locked);
543 while (!(newfe = GetFE())) { /* Get it now, so we don't have to call */
544 /* GetSomeSpace with the host locked, later. This might turn out to */
545 /* have been unneccessary, but that's actually kind of unlikely, since */
546 /* most files are not shared. */
547 GetSomeSpace_r(host, locked);
551 h_Lock_r(host); /* this can yield, so do it before we get any */
553 if (host->hostFlags & HOSTDELETED) {
561 if (type == CB_NORMAL) {
563 TimeCeiling(time(NULL) + TimeOut(fe ? fe->ncbs : 0) +
565 Thead = THead(CBtime(time_out));
566 } else if (type == CB_VOLUME) {
567 time_out = TimeCeiling((60 * 120 + time(NULL)) + ServerBias);
568 Thead = THead(CBtime(time_out));
569 } else if (type == CB_BULK) {
570 /* bulk status can get so many callbacks all at once, and most of them
571 * are probably not for things that will be used for long.
574 TimeCeiling(time(NULL) + ServerBias +
575 TimeOut(22 + (fe ? fe->ncbs : 0)));
576 Thead = THead(CBtime(time_out));
587 fe->volid = fid->Volume;
588 fe->vnode = fid->Vnode;
589 fe->unique = fid->Unique;
592 hash = FEHash(fid->Volume, fid->Unique);
593 fe->fnext = HashTable[hash];
594 HashTable[hash] = fetoi(fe);
596 for (safety = 0, lastcb = cb = itocb(fe->firstcb); cb;
597 lastcb = cb, cb = itocb(cb->cnext), safety++) {
598 if (safety > cbstuff.nblks) {
599 ViceLog(0, ("AddCallBack1: Internal Error -- shutting down.\n"));
600 DumpCallBackState_r();
601 ShutDownAndCore(PANIC);
603 if (cb->hhead == h_htoi(host))
606 if (cb) { /* Already have call back: move to new timeout list */
607 /* don't change delayed callbacks back to normal ones */
608 if (cb->status != CB_DELAYED)
610 /* Only move if new timeout is longer */
611 if (TNorm(ttoi(Thead)) > TNorm(cb->thead)) {
615 if (newfe == NULL) { /* we are using the new FE */
616 fe->firstcb = cbtoi(cb);
618 cb->fhead = fetoi(fe);
623 *(lastcb ? &lastcb->cnext : &fe->firstcb) = cbtoi(cb);
626 cb->fhead = fetoi(fe);
632 /* now free any still-unused callback or host entries */
638 if (!locked) /* freecb and freefe might(?) yield */
641 if (type == CB_NORMAL || type == CB_VOLUME || type == CB_BULK)
642 return time_out - ServerBias; /* Expires sooner at workstation */
648 CompareCBA(const void *e1, const void *e2)
650 const struct cbstruct *cba1 = (const struct cbstruct *)e1;
651 const struct cbstruct *cba2 = (const struct cbstruct *)e2;
652 return ((cba1->hp)->index - (cba2->hp)->index);
655 /* Take an array full of hosts, all held. Break callbacks to them, and
656 * release the holds once you're done.
657 * Currently only works for a single Fid in afidp array.
658 * If you want to make this work with multiple fids, you need to fix
659 * the error handling. One approach would be to force a reset if a
660 * multi-fid call fails, or you could add delayed callbacks for each
661 * fid. You probably also need to sort and remove duplicate hosts.
662 * When this is called from the BreakVolumeCallBacks path, it does NOT
663 * force a reset if the RPC fails, it just marks the host down and tries
664 * to create a delayed callback. */
665 /* N.B. be sure that code works when ncbas == 0 */
666 /* N.B. requires all the cba[*].hp pointers to be valid... */
667 /* This routine does not hold a lock on the host for the duration of
668 * the BreakCallBack RPC, which is a significant deviation from tradition.
669 * It _does_ get a lock on the host before setting VenusDown = 1,
670 * which is sufficient only if VenusDown = 0 only happens when the
671 * lock is held over the RPC and the subsequent VenusDown == 0
672 * wherever that is done. */
674 MultiBreakCallBack_r(struct cbstruct cba[], int ncbas,
675 struct AFSCBFids *afidp)
678 struct rx_connection *conns[MAX_CB_HOSTS];
679 static struct AFSCBs tc = { 0, 0 };
680 int multi_to_cba_map[MAX_CB_HOSTS];
682 opr_Assert(ncbas <= MAX_CB_HOSTS);
684 /* sort cba list to avoid makecall issues */
685 qsort(cba, ncbas, sizeof(struct cbstruct), CompareCBA);
687 /* set up conns for multi-call */
688 for (i = 0, j = 0; i < ncbas; i++) {
689 struct host *thishost = cba[i].hp;
690 if (!thishost || (thishost->hostFlags & HOSTDELETED)) {
693 rx_GetConnection(thishost->callback_rxcon);
694 multi_to_cba_map[j] = i;
695 conns[j++] = thishost->callback_rxcon;
697 rx_SetConnDeadTime(thishost->callback_rxcon, 4);
698 rx_SetConnHardDeadTime(thishost->callback_rxcon, AFS_HARDDEADTIME);
701 if (j) { /* who knows what multi would do with 0 conns? */
705 multi_RXAFSCB_CallBack(afidp, &tc);
711 i = multi_to_cba_map[multi_i];
717 ("BCB: INTERNAL ERROR: hp=%p, cba=%p, thead=%u\n",
721 ** try breaking callbacks on alternate interface addresses
723 if (MultiBreakCallBackAlternateAddress(hp, afidp)) {
726 ("BCB: Failed on file %u.%u.%u, "
727 "Host %p (%s:%d) is down\n",
728 afidp->AFSCBFids_val->Volume,
729 afidp->AFSCBFids_val->Vnode,
730 afidp->AFSCBFids_val->Unique,
732 afs_inet_ntoa_r(hp->host, hoststr),
738 if (!(hp->hostFlags & HOSTDELETED)) {
739 hp->hostFlags |= VENUSDOWN;
741 * We always go into AddCallBack1_r with the host locked
743 AddCallBack1_r(hp, afidp->AFSCBFids_val, itot(idx),
757 for (i = 0; i < ncbas; i++) {
765 /* H_UNLOCK around this so h_FreeConnection does not deadlock.
766 h_FreeConnection should *never* be called on a callback connection,
767 but on 10/27/04 a deadlock occurred where it was, when we know why,
768 this should be reverted. -- shadow */
770 for (i = 0; i < j; i++) {
771 rx_PutConnection(conns[i]);
779 * Break all call backs for fid, except for the specified host (unless flag
780 * is true, in which case all get a callback message. Assumption: the specified
781 * host is h_Held, by the caller; the others aren't.
782 * Specified host may be bogus, that's ok. This used to check to see if the
783 * host was down in two places, once right after the host was h_held, and
784 * again after it was locked. That race condition is incredibly rare and
785 * relatively harmless even when it does occur, so we don't check for it now.
787 /* if flag is true, send a break callback msg to "host", too */
789 BreakCallBack(struct host *xhost, AFSFid * fid, int flag)
791 struct FileEntry *fe;
792 struct CallBack *cb, *nextcb;
793 struct cbstruct cbaDef[MAX_CB_HOSTS], *cba = cbaDef;
794 unsigned int ncbas, cbaAlloc = MAX_CB_HOSTS;
801 ("BCB: BreakCallBack(Host %p all but %s:%d, (%u,%u,%u))\n",
802 xhost, afs_inet_ntoa_r(xhost->host, hoststr), ntohs(xhost->port),
803 fid->Volume, fid->Vnode, fid->Unique));
806 ("BCB: BreakCallBack(No Host, (%u,%u,%u))\n",
807 fid->Volume, fid->Vnode, fid->Unique));
810 cbstuff.BreakCallBacks++;
815 hostindex = xhost ? h_htoi(xhost) : 0;
816 cb = itocb(fe->firstcb);
817 if (!cb || ((fe->ncbs == 1) && (cb->hhead == hostindex) && !flag)) {
818 /* the most common case is what follows the || */
821 tf.AFSCBFids_len = 1;
822 tf.AFSCBFids_val = fid;
824 for (ncbas = 0; cb ; cb = nextcb) {
825 nextcb = itocb(cb->cnext);
826 if ((cb->hhead != hostindex || flag)
827 && (cb->status == CB_BULK || cb->status == CB_NORMAL
828 || cb->status == CB_VOLUME)) {
829 struct host *thishost = h_itoh(cb->hhead);
831 ViceLog(0, ("BCB: BOGUS! cb->hhead is NULL!\n"));
832 } else if (thishost->hostFlags & VENUSDOWN) {
834 ("BCB: %p (%s:%d) is down; delaying break call back\n",
835 thishost, afs_inet_ntoa_r(thishost->host, hoststr),
836 ntohs(thishost->port)));
837 cb->status = CB_DELAYED;
839 if (!(thishost->hostFlags & HOSTDELETED)) {
841 if (ncbas == cbaAlloc) { /* Need more space */
842 int curLen = cbaAlloc*sizeof(cba[0]);
843 struct cbstruct *cbaOld = (cba == cbaDef) ? NULL : cba;
845 /* There are logical contraints elsewhere that the number of hosts
846 (i.e. h_HTSPERBLOCK*h_MAXHOSTTABLES) remains in the realm of a signed "int".
847 cbaAlloc is defined unsigned int hence doubling below cannot overflow
849 cbaAlloc = cbaAlloc<<1; /* double */
850 cba = realloc(cbaOld, cbaAlloc * sizeof(cba[0]));
852 if (cbaOld == NULL) { /* realloc wouldn't have copied from cbaDef */
853 memcpy(cba, cbaDef, curLen);
856 cba[ncbas].hp = thishost;
857 cba[ncbas].thead = cb->thead;
862 CDel(cb, 1); /* Usually first; so this delete
863 * is reasonably inexpensive */
869 struct cbstruct *cba2;
872 for (cba2 = cba, num = ncbas; ncbas > 0; cba2 += num, ncbas -= num) {
873 num = (ncbas > MAX_CB_HOSTS) ? MAX_CB_HOSTS : ncbas;
874 MultiBreakCallBack_r(cba2, num, &tf);
878 if (cba != cbaDef) free(cba);
885 /* Delete (do not break) single call back for fid */
887 DeleteCallBack(struct host *host, AFSFid * fid)
889 struct FileEntry *fe;
894 cbstuff.DeleteCallBacks++;
897 /* do not care if the host has been HOSTDELETED */
903 ("DCB: No call backs for fid (%u, %u, %u)\n", fid->Volume,
904 fid->Vnode, fid->Unique));
907 pcb = FindCBPtr(fe, host);
910 ("DCB: No call back for host %p (%s:%d), (%u, %u, %u)\n",
911 host, afs_inet_ntoa_r(host->host, hoststr), ntohs(host->port),
912 fid->Volume, fid->Vnode, fid->Unique));
926 * Delete (do not break) all call backs for fid. This call doesn't
927 * set all of the various host locks, but it shouldn't really matter
928 * since we're not adding callbacks, but deleting them. I'm not sure
929 * why it doesn't set the lock, however; perhaps it should.
932 DeleteFileCallBacks(AFSFid * fid)
934 struct FileEntry *fe;
940 cbstuff.DeleteFiles++;
945 ("DF: No fid (%u,%u,%u) to delete\n", fid->Volume, fid->Vnode,
949 for (n = 0, cbi = fe->firstcb; cbi; n++) {
962 /* Delete (do not break) all call backs for host. The host should be
965 DeleteAllCallBacks_r(struct host *host, int deletefe)
970 cbstuff.DeleteAllCallBacks++;
971 cbi = first = host->cblist;
973 ViceLog(8, ("DV: no call backs\n"));
981 } while (cbi != first);
987 * Break all delayed call backs for host. Returns 1 if all call backs
988 * successfully broken; 0 otherwise. Assumes host is h_Held and h_Locked.
989 * Must be called with VenusDown set for this host
992 BreakDelayedCallBacks(struct host *host)
996 retVal = BreakDelayedCallBacks_r(host);
1002 BreakDelayedCallBacks_r(struct host *host)
1004 struct AFSFid fids[AFSCBMAX];
1005 int cbi, first, nfids;
1006 struct CallBack *cb;
1009 struct rx_connection *cb_conn;
1011 cbstuff.nbreakers++;
1012 if (!(host->hostFlags & RESETDONE) && !(host->hostFlags & HOSTDELETED)) {
1013 host->hostFlags &= ~ALTADDR; /* alternate addresses are invalid */
1014 cb_conn = host->callback_rxcon;
1015 rx_GetConnection(cb_conn);
1016 if (host->interface) {
1019 RXAFSCB_InitCallBackState3(cb_conn, &FS_HostUUID);
1022 code = RXAFSCB_InitCallBackState(cb_conn);
1024 rx_PutConnection(cb_conn);
1027 host->hostFlags |= ALTADDR; /* alternate addresses are valid */
1031 ("CB: Call back connect back failed (in break delayed) "
1032 "for Host %p (%s:%d)\n",
1033 host, afs_inet_ntoa_r(host->host, hoststr),
1034 ntohs(host->port)));
1036 host->hostFlags |= VENUSDOWN;
1039 ("InitCallBackState success on %p (%s:%d)\n",
1040 host, afs_inet_ntoa_r(host->host, hoststr),
1041 ntohs(host->port)));
1042 /* reset was done successfully */
1043 host->hostFlags |= RESETDONE;
1044 host->hostFlags &= ~VENUSDOWN;
1047 while (!(host->hostFlags & HOSTDELETED)) {
1049 host->hostFlags &= ~VENUSDOWN; /* presume up */
1050 cbi = first = host->cblist;
1054 first = host->cblist;
1057 if (cb->status == CB_DELAYED) {
1058 struct FileEntry *fe = itofe(cb->fhead);
1059 fids[nfids].Volume = fe->volid;
1060 fids[nfids].Vnode = fe->vnode;
1061 fids[nfids].Unique = fe->unique;
1067 } while (cbi && cbi != first && nfids < AFSCBMAX);
1073 if (XCallBackBulk_r(host, fids, nfids)) {
1074 /* Failed, again: put them back, probably with old
1079 ("CB: XCallBackBulk failed, Host %p (%s:%d); "
1080 "callback list follows:\n",
1081 host, afs_inet_ntoa_r(host->host, hoststr),
1082 ntohs(host->port)));
1084 for (i = 0; i < nfids; i++) {
1087 ("CB: Host %p (%s:%d), file %u.%u.%u "
1088 "(part of bulk callback)\n",
1089 host, afs_inet_ntoa_r(host->host, hoststr),
1090 ntohs(host->port), fids[i].Volume,
1091 fids[i].Vnode, fids[i].Unique));
1094 * AddCallBack1_r(host, &fids[i], itot(thead[i]), CB_DELAYED, 1);
1095 * * but it turns out to cause too many tricky locking problems.
1096 * * now, if break delayed fails, screw it. */
1098 host->hostFlags |= VENUSDOWN; /* Failed */
1099 ClearHostCallbacks_r(host, 1 /* locked */ );
1103 if (nfids < AFSCBMAX)
1107 cbstuff.nbreakers--;
1108 /* If we succeeded it's always ok to unset HFE_LATER */
1109 if (!(host->hostFlags & VENUSDOWN))
1110 host->hostFlags &= ~HFE_LATER;
1111 return (host->hostFlags & VENUSDOWN);
1115 MultiBreakVolumeCallBack_r(struct host *host,
1116 struct VCBParams *parms, int deletefe)
1120 if (host->hostFlags & HOSTDELETED)
1123 if (!(host->hostFlags & HCBREAK))
1124 return 0; /* host is not flagged to notify */
1126 if (host->hostFlags & VENUSDOWN) {
1128 /* Do not care if the host is now HOSTDELETED */
1131 ("BVCB: volume callback for Host %p (%s:%d) failed\n",
1132 host, afs_inet_ntoa_r(host->host, hoststr),
1133 ntohs(host->port)));
1135 DeleteAllCallBacks_r(host, deletefe); /* Delete all callback state
1136 * rather than attempting to
1137 * selectively remember to
1138 * delete the volume callbacks
1140 host->hostFlags &= ~(RESETDONE|HCBREAK); /* Do InitCallBackState when host returns */
1144 opr_Assert(parms->ncbas <= MAX_CB_HOSTS);
1146 /* Do not call MultiBreakCallBack on the current host structure
1147 ** because it would prematurely release the hold on the host
1149 if (parms->ncbas == MAX_CB_HOSTS) {
1150 struct AFSCBFids tf;
1152 tf.AFSCBFids_len = 1;
1153 tf.AFSCBFids_val = parms->fid;
1155 /* this releases all the hosts */
1156 MultiBreakCallBack_r(parms->cba, parms->ncbas, &tf);
1160 parms->cba[parms->ncbas].hp = host;
1161 parms->cba[(parms->ncbas)++].thead = parms->thead;
1162 host->hostFlags &= ~HCBREAK;
1164 /* we have more work to do on this host, so make sure we keep a reference
1172 MultiBreakVolumeLaterCallBack(struct host *host, void *rock)
1174 struct VCBParams *parms = (struct VCBParams *)rock;
1177 retval = MultiBreakVolumeCallBack_r(host, parms, 0);
1183 * Break all call backs on a single volume. Don't call this with any
1184 * hosts h_held. Note that this routine clears the callbacks before
1185 * actually breaking them, and that the vnode isn't locked during this
1186 * operation, so that people might see temporary callback loss while
1187 * this function is executing. It is just a temporary state, however,
1188 * since the callback will be broken later by this same function.
1190 * Now uses multi-RX for CallBack RPC in a different thread,
1191 * only marking them here.
1193 extern pthread_cond_t fsync_cond;
1196 BreakVolumeCallBacksLater(VolumeId volume)
1200 struct FileEntry *fe;
1201 struct CallBack *cb;
1205 ViceLog(25, ("Setting later on volume %" AFS_VOLID_FMT "\n",
1206 afs_printable_VolumeId_lu(volume)));
1208 for (hash = 0; hash < FEHASH_SIZE; hash++) {
1209 for (feip = &HashTable[hash]; (fe = itofe(*feip)) != NULL; ) {
1210 if (fe->volid == volume) {
1211 struct CallBack *cbnext;
1212 for (cb = itocb(fe->firstcb); cb; cb = cbnext) {
1213 host = h_itoh(cb->hhead);
1214 host->hostFlags |= HFE_LATER;
1215 cb->status = CB_DELAYED;
1216 cbnext = itocb(cb->cnext);
1219 fe->status |= FE_LATER;
1228 /* didn't find any callbacks, so return right away. */
1232 ViceLog(25, ("Fsync thread wakeup\n"));
1234 opr_cv_broadcast(&fsync_cond);
1240 BreakLaterCallBacks(void)
1245 struct CallBack *cb;
1246 struct FileEntry *fe = NULL;
1247 struct FileEntry *myfe = NULL;
1249 struct VCBParams henumParms;
1250 unsigned short tthead = 0; /* zero is illegal value */
1254 ViceLog(25, ("Looking for FileEntries to unchain\n"));
1257 /* Pick the first volume we see to clean up */
1258 fid.Volume = fid.Vnode = fid.Unique = 0;
1260 for (hash = 0; hash < FEHASH_SIZE; hash++) {
1261 for (feip = &HashTable[hash]; (fe = itofe(*feip)) != NULL; ) {
1262 if (fe && (fe->status & FE_LATER)
1263 && (fid.Volume == 0 || fid.Volume == fe->volid)) {
1264 /* Ugly, but used to avoid left side casting */
1265 struct object *tmpfe;
1267 ("Unchaining for %u:%u:%" AFS_VOLID_FMT "\n", fe->vnode,
1268 fe->unique, afs_printable_VolumeId_lu(fe->volid)));
1269 fid.Volume = fe->volid;
1271 fe->status &= ~FE_LATER; /* not strictly needed */
1272 /* Works since volid is deeper than the largest pointer */
1273 tmpfe = (struct object *)fe;
1274 tmpfe->next = (struct object *)myfe;
1287 /* loop over FEs from myfe and free/break */
1289 for (fe = myfe; fe;) {
1290 struct CallBack *cbnext;
1291 for (cb = itocb(fe->firstcb); cb; cb = cbnext) {
1292 cbnext = itocb(cb->cnext);
1293 host = h_itoh(cb->hhead);
1294 if (cb->status == CB_DELAYED) {
1295 if (!(host->hostFlags & HOSTDELETED)) {
1296 /* mark this host for notification */
1297 host->hostFlags |= HCBREAK;
1298 if (!tthead || (TNorm(tthead) < TNorm(cb->thead))) {
1304 CDel(cb, 0); /* Don't let CDel clean up the fe */
1305 /* leave flag for MultiBreakVolumeCallBack to clear */
1308 ("Found host %p (%s:%d) non-DELAYED cb for %u:%u:%" AFS_VOLID_FMT "\n",
1309 host, afs_inet_ntoa_r(host->host, hoststr),
1310 ntohs(host->port), fe->vnode, fe->unique,
1311 afs_printable_VolumeId_lu(fe->volid)));
1315 fe = (struct FileEntry *)((struct object *)fe)->next;
1320 ViceLog(125, ("Breaking volume %u\n", fid.Volume));
1321 henumParms.ncbas = 0;
1322 henumParms.fid = &fid;
1323 henumParms.thead = tthead;
1325 h_Enumerate(MultiBreakVolumeLaterCallBack, (char *)&henumParms);
1327 if (henumParms.ncbas) { /* do left-overs */
1328 struct AFSCBFids tf;
1329 tf.AFSCBFids_len = 1;
1330 tf.AFSCBFids_val = &fid;
1332 MultiBreakCallBack_r(henumParms.cba, henumParms.ncbas, &tf);
1333 henumParms.ncbas = 0;
1338 /* Arrange to be called again */
1343 * Delete all timed-out call back entries (to be called periodically by file
1347 CleanupTimedOutCallBacks(void)
1350 CleanupTimedOutCallBacks_r();
1356 CleanupTimedOutCallBacks_r(void)
1358 afs_uint32 now = CBtime(time(NULL));
1360 struct CallBack *cb;
1364 while (tfirst <= now) {
1366 cbi = *(thead = THead(tfirst));
1372 ("CCB: deleting timed out call back %x (%s:%d), (%" AFS_VOLID_FMT ",%u,%u)\n",
1373 h_itoh(cb->hhead)->host,
1374 afs_inet_ntoa_r(h_itoh(cb->hhead)->host, hoststr),
1375 h_itoh(cb->hhead)->port,
1376 afs_printable_VolumeId_lu(itofe(cb->fhead)->volid),
1377 itofe(cb->fhead)->vnode, itofe(cb->fhead)->unique));
1381 if (ntimedout > cbstuff.nblks) {
1382 ViceLog(0, ("CCB: Internal Error -- shutting down...\n"));
1383 DumpCallBackState_r();
1384 ShutDownAndCore(PANIC);
1386 } while (cbi != *thead);
1391 cbstuff.CBsTimedOut += ntimedout;
1392 ViceLog(7, ("CCB: deleted %d timed out callbacks\n", ntimedout));
1393 return (ntimedout > 0);
1397 * parameters to pass to lih*_r from h_Enumerate_r when trying to find a host
1398 * from which to clear callbacks.
1402 * Points to the least interesting host found; try to clear callbacks on
1403 * this host after h_Enumerate_r(lih*_r)'ing.
1408 * The last host we got from lih*_r, but we couldn't clear its callbacks
1409 * for some reason. Choose the next-best host after this one (with the
1410 * current lih*_r, this means to only select hosts that have an ActiveCall
1411 * newer than lastlih).
1413 struct host *lastlih;
1416 /* Value of host->refCount that allows us to reliably infer that
1417 * host may be held by some other thread */
1418 #define OTHER_MUSTHOLD_LIH 2
1420 /* This version does not allow 'host' to be selected unless its ActiveCall
1421 * is newer than 'params->lastlih' which is the host with the oldest
1422 * ActiveCall from the last pass (if it is provided). We filter out any hosts
1423 * that are are held by other threads.
1425 * There is a small problem here, but it may not be easily fixable. Say we
1426 * select some host A, and give it back to GetSomeSpace_r. GSS_r for some
1427 * reason cannot clear the callbacks on A, and so calls us again with
1428 * lastlih = A. Suppose there is another host B that has the same ActiveCall
1429 * time as A. We will now skip over host B, since
1430 * 'hostB->ActiveCall > hostA->ActiveCall' is not true. This could result in
1431 * us prematurely going to the GSS_r 2nd or 3rd pass, and making us a little
1432 * inefficient. This should be pretty rare, though, except perhaps in cases
1433 * with very small numbers of hosts.
1435 * Also filter out any hosts with HOSTDELETED set. h_Enumerate_r should in
1436 * theory not give these to us anyway, but be paranoid.
1439 lih0_r(struct host *host, void *rock)
1441 struct lih_params *params = (struct lih_params *)rock;
1443 /* OTHER_MUSTHOLD_LIH is because the h_Enum loop holds us once */
1445 && (!(host->hostFlags & HOSTDELETED))
1446 && (host->refCount < OTHER_MUSTHOLD_LIH)
1447 && (!params->lih || host->ActiveCall < params->lih->ActiveCall)
1448 && (!params->lastlih || host->ActiveCall > params->lastlih->ActiveCall)) {
1451 h_Release_r(params->lih); /* release prev host */
1460 /* same as lih0_r, except we do not prevent held hosts from being selected. */
1462 lih1_r(struct host *host, void *rock)
1464 struct lih_params *params = (struct lih_params *)rock;
1467 && (!(host->hostFlags & HOSTDELETED))
1468 && (!params->lih || host->ActiveCall < params->lih->ActiveCall)
1469 && (!params->lastlih || host->ActiveCall > params->lastlih->ActiveCall)) {
1472 h_Release_r(params->lih); /* release prev host */
1481 /* This could be upgraded to get more space each time */
1482 /* first pass: sequentially find the oldest host which isn't held by
1483 anyone for which we can clear callbacks;
1485 /* second pass: sequentially find the oldest host regardless of
1486 whether or not the host is held; skipping 'hostp' */
1487 /* third pass: attempt to clear callbacks from 'hostp' */
1488 /* always called with hostp unlocked */
1490 /* Note: hostlist is ordered most recently created host first and
1491 * its order has no relationship to the most recently used. */
1492 extern struct host *hostList;
1494 GetSomeSpace_r(struct host *hostp, int locked)
1497 struct lih_params params;
1500 if (cbstuff.GotSomeSpaces == 0) {
1501 /* only log this once; if GSS is getting called constantly, that's not
1502 * good but don't make things worse by spamming the log. */
1503 ViceLog(0, ("We have run out of callback space; forcing callback revocation. "
1504 "This suggests the fileserver is configured with insufficient "
1505 "callbacks; you probably want to increase the -cb fileserver "
1506 "parameter (current setting: %u). The fileserver will continue "
1507 "to operate, but this may indicate a severe performance problem\n",
1509 ViceLog(0, ("This message is logged at most once; for more information "
1510 "see the OpenAFS documentation and fileserver xstat collection 3\n"));
1513 cbstuff.GotSomeSpaces++;
1515 ("GSS: First looking for timed out call backs via CleanupCallBacks\n"));
1516 if (CleanupTimedOutCallBacks_r()) {
1522 params.lastlih = NULL;
1527 h_Enumerate_r(i == 0 ? lih0_r : lih1_r, hostList, ¶ms);
1530 if (params.lastlih) {
1531 h_Release_r(params.lastlih);
1532 params.lastlih = NULL;
1536 /* note that 'hp' was held by lih*_r; we will need to release it */
1538 if ((hp != hostp) && !ClearHostCallbacks_r(hp, 0 /* not locked or held */ )) {
1543 params.lastlih = hp;
1544 /* params.lastlih will be released on the next iteration, after
1549 * Next time try getting callbacks from any host even if
1550 * it's held, since the only other option is starvation for
1551 * the file server (i.e. until the callback timeout arrives).
1554 params.lastlih = NULL;
1557 ("GSS: Try harder for longest inactive host cnt= %d\n",
1562 /* Could not obtain space from other hosts, clear hostp's callback state */
1567 ClearHostCallbacks_r(hostp, 1 /*already locked */ );
1574 /* locked - set if caller has already locked the host */
1576 ClearHostCallbacks_r(struct host *hp, int locked)
1580 struct rx_connection *cb_conn = NULL;
1583 ("GSS: Delete longest inactive host %p (%s:%d)\n",
1584 hp, afs_inet_ntoa_r(hp->host, hoststr), ntohs(hp->port)));
1586 if ((hp->hostFlags & HOSTDELETED)) {
1587 /* hp could go away after reacquiring H_LOCK in h_NBLock_r, so we can't
1588 * really use it; its callbacks will get cleared anyway when
1589 * h_TossStuff_r gets its hands on it */
1595 /** Try a non-blocking lock. If the lock is already held return
1596 * after releasing hold on hp
1599 if (h_NBLock_r(hp)) {
1604 if (hp->Console & 2) {
1606 * If the special console field is set it means that a thread
1607 * is waiting in AddCallBack1 after it set pointers to the
1608 * file entry and/or callback entry. Because of the bogus
1609 * usage of h_hold it won't prevent from another thread, this
1610 * one, to remove all the callbacks so just to be safe we keep
1611 * a reference. NOTE, on the last phase we'll free the calling
1612 * host's callbacks but that's ok...
1616 DeleteAllCallBacks_r(hp, 1);
1617 if (hp->hostFlags & VENUSDOWN) {
1618 hp->hostFlags &= ~RESETDONE; /* remember that we must do a reset */
1619 } else if (!(hp->hostFlags & HOSTDELETED)) {
1620 /* host is up, try a call */
1621 hp->hostFlags &= ~ALTADDR; /* alternate addresses are invalid */
1622 cb_conn = hp->callback_rxcon;
1623 rx_GetConnection(hp->callback_rxcon);
1624 if (hp->interface) {
1627 RXAFSCB_InitCallBackState3(cb_conn, &FS_HostUUID);
1630 code = RXAFSCB_InitCallBackState(cb_conn);
1632 rx_PutConnection(cb_conn);
1635 hp->hostFlags |= ALTADDR; /* alternate addresses are valid */
1637 /* failed, mark host down and need reset */
1638 hp->hostFlags |= VENUSDOWN;
1639 hp->hostFlags &= ~RESETDONE;
1641 /* reset succeeded, we're done */
1642 hp->hostFlags |= RESETDONE;
1651 #endif /* INTERPRET_DUMP */
1655 PrintCallBackStats(void)
1658 "%d add CB, %d break CB, %d del CB, %d del FE, %d CB's timed out, %d space reclaim, %d del host\n",
1659 cbstuff.AddCallBacks, cbstuff.BreakCallBacks,
1660 cbstuff.DeleteCallBacks, cbstuff.DeleteFiles, cbstuff.CBsTimedOut,
1661 cbstuff.GotSomeSpaces, cbstuff.DeleteAllCallBacks);
1662 fprintf(stderr, "%d CBs, %d FEs, (%d of total of %d 16-byte blocks)\n",
1663 cbstuff.nCBs, cbstuff.nFEs, cbstuff.nCBs + cbstuff.nFEs,
1665 fprintf(stderr, "%d GSS1, %d GSS2, %d GSS3, %d GSS4, %d GSS5 (internal counters)\n",
1666 cbstuff.GSS1, cbstuff.GSS2, cbstuff.GSS3, cbstuff.GSS4, cbstuff.GSS5);
1671 #define MAGIC 0x12345678 /* To check byte ordering of dump when it is read in */
1672 #define MAGICV2 0x12345679 /* To check byte ordering & version of dump when it is read in */
1675 #ifndef INTERPRET_DUMP
1677 #ifdef AFS_DEMAND_ATTACH_FS
1680 * callback state serialization
1682 static int cb_stateSaveTimeouts(struct fs_dump_state * state);
1683 static int cb_stateSaveFEHash(struct fs_dump_state * state);
1684 static int cb_stateSaveFEs(struct fs_dump_state * state);
1685 static int cb_stateSaveFE(struct fs_dump_state * state, struct FileEntry * fe);
1686 static int cb_stateRestoreTimeouts(struct fs_dump_state * state);
1687 static int cb_stateRestoreFEHash(struct fs_dump_state * state);
1688 static int cb_stateRestoreFEs(struct fs_dump_state * state);
1689 static int cb_stateRestoreFE(struct fs_dump_state * state);
1690 static int cb_stateRestoreCBs(struct fs_dump_state * state, struct FileEntry * fe,
1691 struct iovec * iov, int niovecs);
1693 static int cb_stateVerifyFEHash(struct fs_dump_state * state);
1694 static int cb_stateVerifyFE(struct fs_dump_state * state, struct FileEntry * fe);
1695 static int cb_stateVerifyFCBList(struct fs_dump_state * state, struct FileEntry * fe);
1696 static int cb_stateVerifyTimeoutQueues(struct fs_dump_state * state);
1698 static int cb_stateFEToDiskEntry(struct FileEntry *, struct FEDiskEntry *);
1699 static int cb_stateDiskEntryToFE(struct fs_dump_state * state,
1700 struct FEDiskEntry *, struct FileEntry *);
1702 static int cb_stateCBToDiskEntry(struct CallBack *, struct CBDiskEntry *);
1703 static int cb_stateDiskEntryToCB(struct fs_dump_state * state,
1704 struct CBDiskEntry *, struct CallBack *);
1706 static int cb_stateFillHeader(struct callback_state_header * hdr);
1707 static int cb_stateCheckHeader(struct callback_state_header * hdr);
1709 static int cb_stateAllocMap(struct fs_dump_state * state);
1712 cb_stateSave(struct fs_dump_state * state)
1716 AssignInt64(state->eof_offset, &state->hdr->cb_offset);
1718 /* invalidate callback state header */
1719 memset(state->cb_hdr, 0, sizeof(struct callback_state_header));
1720 if (fs_stateWriteHeader(state, &state->hdr->cb_offset, state->cb_hdr,
1721 sizeof(struct callback_state_header))) {
1726 fs_stateIncEOF(state, sizeof(struct callback_state_header));
1728 /* dump timeout state */
1729 if (cb_stateSaveTimeouts(state)) {
1734 /* dump fe hashtable state */
1735 if (cb_stateSaveFEHash(state)) {
1740 /* dump callback state */
1741 if (cb_stateSaveFEs(state)) {
1746 /* write the callback state header to disk */
1747 cb_stateFillHeader(state->cb_hdr);
1748 if (fs_stateWriteHeader(state, &state->hdr->cb_offset, state->cb_hdr,
1749 sizeof(struct callback_state_header))) {
1759 cb_stateRestore(struct fs_dump_state * state)
1763 if (fs_stateReadHeader(state, &state->hdr->cb_offset, state->cb_hdr,
1764 sizeof(struct callback_state_header))) {
1769 if (cb_stateCheckHeader(state->cb_hdr)) {
1774 if (cb_stateAllocMap(state)) {
1779 if (cb_stateRestoreTimeouts(state)) {
1784 if (cb_stateRestoreFEHash(state)) {
1789 /* restore FEs and CBs from disk */
1790 if (cb_stateRestoreFEs(state)) {
1795 /* restore the timeout queue heads */
1796 tfirst = state->cb_hdr->tfirst;
1803 cb_stateRestoreIndices(struct fs_dump_state * state)
1806 struct FileEntry * fe;
1807 struct CallBack * cb;
1809 /* restore indices in the FileEntry structures */
1810 for (i = 1; i < state->fe_map.len; i++) {
1811 if (state->fe_map.entries[i].new_idx) {
1812 fe = itofe(state->fe_map.entries[i].new_idx);
1814 /* restore the fe->fnext entry */
1815 if (fe_OldToNew(state, fe->fnext, &fe->fnext)) {
1820 /* restore the fe->firstcb entry */
1821 if (cb_OldToNew(state, fe->firstcb, &fe->firstcb)) {
1828 /* restore indices in the CallBack structures */
1829 for (i = 1; i < state->cb_map.len; i++) {
1830 if (state->cb_map.entries[i].new_idx) {
1831 cb = itocb(state->cb_map.entries[i].new_idx);
1833 /* restore the cb->cnext entry */
1834 if (cb_OldToNew(state, cb->cnext, &cb->cnext)) {
1839 /* restore the cb->fhead entry */
1840 if (fe_OldToNew(state, cb->fhead, &cb->fhead)) {
1845 /* restore the cb->hhead entry */
1846 if (h_OldToNew(state, cb->hhead, &cb->hhead)) {
1851 /* restore the cb->tprev entry */
1852 if (cb_OldToNew(state, cb->tprev, &cb->tprev)) {
1857 /* restore the cb->tnext entry */
1858 if (cb_OldToNew(state, cb->tnext, &cb->tnext)) {
1863 /* restore the cb->hprev entry */
1864 if (cb_OldToNew(state, cb->hprev, &cb->hprev)) {
1869 /* restore the cb->hnext entry */
1870 if (cb_OldToNew(state, cb->hnext, &cb->hnext)) {
1877 /* restore the timeout queue head indices */
1878 for (i = 0; i < state->cb_timeout_hdr->records; i++) {
1879 if (cb_OldToNew(state, timeout[i], &timeout[i])) {
1885 /* restore the FE hash table queue heads */
1886 for (i = 0; i < state->cb_fehash_hdr->records; i++) {
1887 if (fe_OldToNew(state, HashTable[i], &HashTable[i])) {
1898 cb_stateVerify(struct fs_dump_state * state)
1902 if (cb_stateVerifyFEHash(state)) {
1906 if (cb_stateVerifyTimeoutQueues(state)) {
1914 cb_stateVerifyFEHash(struct fs_dump_state * state)
1917 struct FileEntry * fe;
1918 afs_uint32 fei, chain_len;
1920 for (i = 0; i < FEHASH_SIZE; i++) {
1922 for (fei = HashTable[i], fe = itofe(fei);
1924 fei = fe->fnext, fe = itofe(fei)) {
1925 if (fei > cbstuff.nblks) {
1926 ViceLog(0, ("cb_stateVerifyFEHash: error: index out of range (fei=%d)\n", fei));
1930 if (cb_stateVerifyFE(state, fe)) {
1933 if (chain_len > FS_STATE_FE_MAX_HASH_CHAIN_LEN) {
1934 ViceLog(0, ("cb_stateVerifyFEHash: error: hash chain %d length exceeds %d; assuming there's a loop\n",
1935 i, FS_STATE_FE_MAX_HASH_CHAIN_LEN));
1947 cb_stateVerifyFE(struct fs_dump_state * state, struct FileEntry * fe)
1951 if ((fe->firstcb && !fe->ncbs) ||
1952 (!fe->firstcb && fe->ncbs)) {
1953 ViceLog(0, ("cb_stateVerifyFE: error: fe->firstcb does not agree with fe->ncbs (fei=%lu, fe->firstcb=%lu, fe->ncbs=%lu)\n",
1954 afs_printable_uint32_lu(fetoi(fe)),
1955 afs_printable_uint32_lu(fe->firstcb),
1956 afs_printable_uint32_lu(fe->ncbs)));
1959 if (cb_stateVerifyFCBList(state, fe)) {
1960 ViceLog(0, ("cb_stateVerifyFE: error: FCBList failed verification (fei=%lu)\n",
1961 afs_printable_uint32_lu(fetoi(fe))));
1969 cb_stateVerifyFCBList(struct fs_dump_state * state, struct FileEntry * fe)
1972 afs_uint32 cbi, fei, chain_len = 0;
1973 struct CallBack * cb;
1977 for (cbi = fe->firstcb, cb = itocb(cbi);
1979 cbi = cb->cnext, cb = itocb(cbi)) {
1980 if (cbi > cbstuff.nblks) {
1981 ViceLog(0, ("cb_stateVerifyFCBList: error: list index out of range (cbi=%d, ncbs=%d)\n",
1982 cbi, cbstuff.nblks));
1986 if (cb->fhead != fei) {
1987 ViceLog(0, ("cb_stateVerifyFCBList: error: cb->fhead != fei (fei=%d, cb->fhead=%d)\n",
1991 if (chain_len > FS_STATE_FCB_MAX_LIST_LEN) {
1992 ViceLog(0, ("cb_stateVerifyFCBList: error: list length exceeds %d (fei=%d); assuming there's a loop\n",
1993 FS_STATE_FCB_MAX_LIST_LEN, fei));
2000 if (fe->ncbs != chain_len) {
2001 ViceLog(0, ("cb_stateVerifyFCBList: error: list length mismatch (len=%d, fe->ncbs=%d)\n",
2002 chain_len, fe->ncbs));
2011 cb_stateVerifyHCBList(struct fs_dump_state * state, struct host * host)
2014 afs_uint32 hi, chain_len, cbi;
2015 struct CallBack *cb, *ncb;
2020 for (cbi = host->cblist, cb = itocb(cbi);
2022 cbi = cb->hnext, cb = ncb) {
2023 if (chain_len && (host->cblist == cbi)) {
2024 /* we've wrapped around the circular list, and everything looks ok */
2027 if (cb->hhead != hi) {
2028 ViceLog(0, ("cb_stateVerifyHCBList: error: incorrect cb->hhead (cbi=%d, h->index=%d, cb->hhead=%d)\n",
2029 cbi, hi, cb->hhead));
2032 if (!cb->hprev || !cb->hnext) {
2033 ViceLog(0, ("cb_stateVerifyHCBList: error: null index in circular list (cbi=%d, h->index=%d)\n",
2038 if ((cb->hprev > cbstuff.nblks) ||
2039 (cb->hnext > cbstuff.nblks)) {
2040 ViceLog(0, ("cb_stateVerifyHCBList: error: list index out of range (cbi=%d, h->index=%d, cb->hprev=%d, cb->hnext=%d, nCBs=%d)\n",
2041 cbi, hi, cb->hprev, cb->hnext, cbstuff.nblks));
2045 ncb = itocb(cb->hnext);
2046 if (cbi != ncb->hprev) {
2047 ViceLog(0, ("cb_stateVerifyHCBList: error: corrupt linked list (cbi=%d, h->index=%d)\n",
2052 if (chain_len > FS_STATE_HCB_MAX_LIST_LEN) {
2053 ViceLog(0, ("cb_stateVerifyFCBList: error: list length exceeds %d (h->index=%d); assuming there's a loop\n",
2054 FS_STATE_HCB_MAX_LIST_LEN, hi));
2066 cb_stateVerifyTimeoutQueues(struct fs_dump_state * state)
2069 afs_uint32 cbi, chain_len;
2070 struct CallBack *cb, *ncb;
2072 for (i = 0; i < CB_NUM_TIMEOUT_QUEUES; i++) {
2074 for (cbi = timeout[i], cb = itocb(cbi);
2076 cbi = cb->tnext, cb = ncb) {
2077 if (chain_len && (cbi == timeout[i])) {
2078 /* we've wrapped around the circular list, and everything looks ok */
2081 if (cbi > cbstuff.nblks) {
2082 ViceLog(0, ("cb_stateVerifyTimeoutQueues: error: list index out of range (cbi=%d, tindex=%d)\n",
2087 if (itot(cb->thead) != &timeout[i]) {
2088 ViceLog(0, ("cb_stateVerifyTimeoutQueues: error: cb->thead points to wrong timeout queue (tindex=%d, cbi=%d, cb->thead=%d)\n",
2089 i, cbi, cb->thead));
2092 if (!cb->tprev || !cb->tnext) {
2093 ViceLog(0, ("cb_stateVerifyTimeoutQueues: null index in circular list (cbi=%d, tindex=%d)\n",
2098 if ((cb->tprev > cbstuff.nblks) ||
2099 (cb->tnext > cbstuff.nblks)) {
2100 ViceLog(0, ("cb_stateVerifyTimeoutQueues: list index out of range (cbi=%d, tindex=%d, cb->tprev=%d, cb->tnext=%d, nCBs=%d)\n",
2101 cbi, i, cb->tprev, cb->tnext, cbstuff.nblks));
2105 ncb = itocb(cb->tnext);
2106 if (cbi != ncb->tprev) {
2107 ViceLog(0, ("cb_stateVerifyTimeoutQueues: corrupt linked list (cbi=%d, tindex=%d)\n",
2112 if (chain_len > FS_STATE_TCB_MAX_LIST_LEN) {
2113 ViceLog(0, ("cb_stateVerifyTimeoutQueues: list length exceeds %d (tindex=%d); assuming there's a loop\n",
2114 FS_STATE_TCB_MAX_LIST_LEN, i));
2126 cb_stateSaveTimeouts(struct fs_dump_state * state)
2129 struct iovec iov[2];
2131 AssignInt64(state->eof_offset, &state->cb_hdr->timeout_offset);
2133 memset(state->cb_timeout_hdr, 0, sizeof(struct callback_state_fehash_header));
2134 state->cb_timeout_hdr->magic = CALLBACK_STATE_TIMEOUT_MAGIC;
2135 state->cb_timeout_hdr->records = CB_NUM_TIMEOUT_QUEUES;
2136 state->cb_timeout_hdr->len = sizeof(struct callback_state_timeout_header) +
2137 (state->cb_timeout_hdr->records * sizeof(afs_uint32));
2139 iov[0].iov_base = (char *)state->cb_timeout_hdr;
2140 iov[0].iov_len = sizeof(struct callback_state_timeout_header);
2141 iov[1].iov_base = (char *)timeout;
2142 iov[1].iov_len = sizeof(timeout);
2144 if (fs_stateSeek(state, &state->cb_hdr->timeout_offset)) {
2149 if (fs_stateWriteV(state, iov, 2)) {
2154 fs_stateIncEOF(state, state->cb_timeout_hdr->len);
2161 cb_stateRestoreTimeouts(struct fs_dump_state * state)
2165 if (fs_stateReadHeader(state, &state->cb_hdr->timeout_offset,
2166 state->cb_timeout_hdr,
2167 sizeof(struct callback_state_timeout_header))) {
2172 if (state->cb_timeout_hdr->magic != CALLBACK_STATE_TIMEOUT_MAGIC) {
2176 if (state->cb_timeout_hdr->records != CB_NUM_TIMEOUT_QUEUES) {
2181 len = state->cb_timeout_hdr->records * sizeof(afs_uint32);
2183 if (state->cb_timeout_hdr->len !=
2184 (sizeof(struct callback_state_timeout_header) + len)) {
2189 if (fs_stateRead(state, timeout, len)) {
2199 cb_stateSaveFEHash(struct fs_dump_state * state)
2202 struct iovec iov[2];
2204 AssignInt64(state->eof_offset, &state->cb_hdr->fehash_offset);
2206 memset(state->cb_fehash_hdr, 0, sizeof(struct callback_state_fehash_header));
2207 state->cb_fehash_hdr->magic = CALLBACK_STATE_FEHASH_MAGIC;
2208 state->cb_fehash_hdr->records = FEHASH_SIZE;
2209 state->cb_fehash_hdr->len = sizeof(struct callback_state_fehash_header) +
2210 (state->cb_fehash_hdr->records * sizeof(afs_uint32));
2212 iov[0].iov_base = (char *)state->cb_fehash_hdr;
2213 iov[0].iov_len = sizeof(struct callback_state_fehash_header);
2214 iov[1].iov_base = (char *)HashTable;
2215 iov[1].iov_len = sizeof(HashTable);
2217 if (fs_stateSeek(state, &state->cb_hdr->fehash_offset)) {
2222 if (fs_stateWriteV(state, iov, 2)) {
2227 fs_stateIncEOF(state, state->cb_fehash_hdr->len);
2234 cb_stateRestoreFEHash(struct fs_dump_state * state)
2238 if (fs_stateReadHeader(state, &state->cb_hdr->fehash_offset,
2239 state->cb_fehash_hdr,
2240 sizeof(struct callback_state_fehash_header))) {
2245 if (state->cb_fehash_hdr->magic != CALLBACK_STATE_FEHASH_MAGIC) {
2249 if (state->cb_fehash_hdr->records != FEHASH_SIZE) {
2254 len = state->cb_fehash_hdr->records * sizeof(afs_uint32);
2256 if (state->cb_fehash_hdr->len !=
2257 (sizeof(struct callback_state_fehash_header) + len)) {
2262 if (fs_stateRead(state, HashTable, len)) {
2272 cb_stateSaveFEs(struct fs_dump_state * state)
2276 struct FileEntry *fe;
2278 AssignInt64(state->eof_offset, &state->cb_hdr->fe_offset);
2280 for (hash = 0; hash < FEHASH_SIZE ; hash++) {
2281 for (fei = HashTable[hash]; fei; fei = fe->fnext) {
2283 if (cb_stateSaveFE(state, fe)) {
2295 cb_stateRestoreFEs(struct fs_dump_state * state)
2297 int count, nFEs, ret = 0;
2299 nFEs = state->cb_hdr->nFEs;
2301 for (count = 0; count < nFEs; count++) {
2302 if (cb_stateRestoreFE(state)) {
2313 cb_stateSaveFE(struct fs_dump_state * state, struct FileEntry * fe)
2315 int ret = 0, iovcnt, cbi, written = 0;
2317 struct callback_state_entry_header hdr;
2318 struct FEDiskEntry fedsk;
2319 struct CBDiskEntry cbdsk[16];
2320 struct iovec iov[16];
2321 struct CallBack *cb;
2324 if (fei > state->cb_hdr->fe_max) {
2325 state->cb_hdr->fe_max = fei;
2328 memset(&hdr, 0, sizeof(struct callback_state_entry_header));
2330 if (cb_stateFEToDiskEntry(fe, &fedsk)) {
2335 iov[0].iov_base = (char *)&hdr;
2336 iov[0].iov_len = sizeof(hdr);
2337 iov[1].iov_base = (char *)&fedsk;
2338 iov[1].iov_len = sizeof(struct FEDiskEntry);
2341 for (cbi = fe->firstcb, cb = itocb(cbi);
2343 cbi = cb->cnext, cb = itocb(cbi), hdr.nCBs++) {
2344 if (cbi > state->cb_hdr->cb_max) {
2345 state->cb_hdr->cb_max = cbi;
2347 if (cb_stateCBToDiskEntry(cb, &cbdsk[iovcnt])) {
2351 cbdsk[iovcnt].index = cbi;
2352 iov[iovcnt].iov_base = (char *)&cbdsk[iovcnt];
2353 iov[iovcnt].iov_len = sizeof(struct CBDiskEntry);
2355 if ((iovcnt == 16) || (!cb->cnext)) {
2356 if (fs_stateWriteV(state, iov, iovcnt)) {
2365 hdr.magic = CALLBACK_STATE_ENTRY_MAGIC;
2366 hdr.len = sizeof(hdr) + sizeof(struct FEDiskEntry) +
2367 (hdr.nCBs * sizeof(struct CBDiskEntry));
2370 if (fs_stateWriteV(state, iov, iovcnt)) {
2375 if (fs_stateWriteHeader(state, &state->eof_offset, &hdr, sizeof(hdr))) {
2381 fs_stateIncEOF(state, hdr.len);
2384 if (fs_stateSeek(state, &state->eof_offset)) {
2390 state->cb_hdr->nFEs++;
2391 state->cb_hdr->nCBs += hdr.nCBs;
2398 cb_stateRestoreFE(struct fs_dump_state * state)
2400 int ret = 0, iovcnt, nCBs;
2401 struct callback_state_entry_header hdr;
2402 struct FEDiskEntry fedsk;
2403 struct CBDiskEntry cbdsk[16];
2404 struct iovec iov[16];
2405 struct FileEntry * fe;
2407 iov[0].iov_base = (char *)&hdr;
2408 iov[0].iov_len = sizeof(hdr);
2409 iov[1].iov_base = (char *)&fedsk;
2410 iov[1].iov_len = sizeof(fedsk);
2413 if (fs_stateReadV(state, iov, iovcnt)) {
2418 if (hdr.magic != CALLBACK_STATE_ENTRY_MAGIC) {
2425 ViceLog(0, ("cb_stateRestoreFE: ran out of free FileEntry structures\n"));
2430 if (cb_stateDiskEntryToFE(state, &fedsk, fe)) {
2436 for (iovcnt = 0, nCBs = 0;
2439 iov[iovcnt].iov_base = (char *)&cbdsk[iovcnt];
2440 iov[iovcnt].iov_len = sizeof(struct CBDiskEntry);
2442 if ((iovcnt == 16) || (nCBs == hdr.nCBs - 1)) {
2443 if (fs_stateReadV(state, iov, iovcnt)) {
2447 if (cb_stateRestoreCBs(state, fe, iov, iovcnt)) {
2461 cb_stateRestoreCBs(struct fs_dump_state * state, struct FileEntry * fe,
2462 struct iovec * iov, int niovecs)
2465 struct CallBack * cb;
2466 struct CBDiskEntry * cbdsk;
2468 for (idx = 0; idx < niovecs; idx++) {
2469 cbdsk = (struct CBDiskEntry *) iov[idx].iov_base;
2471 if (cbdsk->cb.hhead < state->h_map.len &&
2472 state->h_map.entries[cbdsk->cb.hhead].valid == FS_STATE_IDX_SKIPPED) {
2476 if ((cb = GetCB()) == NULL) {
2477 ViceLog(0, ("cb_stateRestoreCBs: ran out of free CallBack structures\n"));
2481 if (cb_stateDiskEntryToCB(state, cbdsk, cb)) {
2482 ViceLog(0, ("cb_stateRestoreCBs: corrupt CallBack disk entry\n"));
2494 cb_stateFillHeader(struct callback_state_header * hdr)
2496 hdr->stamp.magic = CALLBACK_STATE_MAGIC;
2497 hdr->stamp.version = CALLBACK_STATE_VERSION;
2498 hdr->tfirst = tfirst;
2503 cb_stateCheckHeader(struct callback_state_header * hdr)
2507 if (hdr->stamp.magic != CALLBACK_STATE_MAGIC) {
2509 } else if (hdr->stamp.version != CALLBACK_STATE_VERSION) {
2511 } else if ((hdr->nFEs > cbstuff.nblks) || (hdr->nCBs > cbstuff.nblks)) {
2512 ViceLog(0, ("cb_stateCheckHeader: saved callback state larger than callback memory allocation\n"));
2518 /* disk entry conversion routines */
2520 cb_stateFEToDiskEntry(struct FileEntry * in, struct FEDiskEntry * out)
2522 memcpy(&out->fe, in, sizeof(struct FileEntry));
2523 out->index = fetoi(in);
2528 cb_stateDiskEntryToFE(struct fs_dump_state * state,
2529 struct FEDiskEntry * in, struct FileEntry * out)
2533 memcpy(out, &in->fe, sizeof(struct FileEntry));
2535 /* setup FE map entry */
2536 if (!in->index || (in->index >= state->fe_map.len)) {
2537 ViceLog(0, ("cb_stateDiskEntryToFE: index (%d) out of range",
2542 state->fe_map.entries[in->index].valid = FS_STATE_IDX_VALID;
2543 state->fe_map.entries[in->index].old_idx = in->index;
2544 state->fe_map.entries[in->index].new_idx = fetoi(out);
2551 cb_stateCBToDiskEntry(struct CallBack * in, struct CBDiskEntry * out)
2553 memcpy(&out->cb, in, sizeof(struct CallBack));
2554 out->index = cbtoi(in);
2559 cb_stateDiskEntryToCB(struct fs_dump_state * state,
2560 struct CBDiskEntry * in, struct CallBack * out)
2564 memcpy(out, &in->cb, sizeof(struct CallBack));
2566 /* setup CB map entry */
2567 if (!in->index || (in->index >= state->cb_map.len)) {
2568 ViceLog(0, ("cb_stateDiskEntryToCB: index (%d) out of range\n",
2573 state->cb_map.entries[in->index].valid = FS_STATE_IDX_VALID;
2574 state->cb_map.entries[in->index].old_idx = in->index;
2575 state->cb_map.entries[in->index].new_idx = cbtoi(out);
2581 /* index map routines */
2583 cb_stateAllocMap(struct fs_dump_state * state)
2585 state->fe_map.len = state->cb_hdr->fe_max + 1;
2586 state->cb_map.len = state->cb_hdr->cb_max + 1;
2587 state->fe_map.entries = (struct idx_map_entry_t *)
2588 calloc(state->fe_map.len, sizeof(struct idx_map_entry_t));
2589 state->cb_map.entries = (struct idx_map_entry_t *)
2590 calloc(state->cb_map.len, sizeof(struct idx_map_entry_t));
2591 return ((state->fe_map.entries != NULL) && (state->cb_map.entries != NULL)) ? 0 : 1;
2595 fe_OldToNew(struct fs_dump_state * state, afs_uint32 old, afs_uint32 * new)
2599 /* FEs use a one-based indexing system, so old==0 implies no mapping */
2605 if (old >= state->fe_map.len) {
2606 ViceLog(0, ("fe_OldToNew: index %d is out of range\n", old));
2608 } else if (state->fe_map.entries[old].valid != FS_STATE_IDX_VALID ||
2609 state->fe_map.entries[old].old_idx != old) { /* sanity check */
2610 ViceLog(0, ("fe_OldToNew: index %d points to an invalid FileEntry record\n", old));
2613 *new = state->fe_map.entries[old].new_idx;
2621 cb_OldToNew(struct fs_dump_state * state, afs_uint32 old, afs_uint32 * new)
2625 /* CBs use a one-based indexing system, so old==0 implies no mapping */
2631 if (old >= state->cb_map.len) {
2632 ViceLog(0, ("cb_OldToNew: index %d is out of range\n", old));
2634 } else if (state->cb_map.entries[old].valid != FS_STATE_IDX_VALID ||
2635 state->cb_map.entries[old].old_idx != old) { /* sanity check */
2636 ViceLog(0, ("cb_OldToNew: index %d points to an invalid CallBack record\n", old));
2639 *new = state->cb_map.entries[old].new_idx;
2645 #endif /* AFS_DEMAND_ATTACH_FS */
2648 DumpCallBackState_r(void)
2651 afs_uint32 magic = MAGICV2, now = (afs_int32) time(NULL), freelisthead;
2653 oflag = O_WRONLY | O_CREAT | O_TRUNC;
2657 fd = open(AFSDIR_SERVER_CBKDUMP_FILEPATH, oflag, 0666);
2660 ("Couldn't create callback dump file %s\n",
2661 AFSDIR_SERVER_CBKDUMP_FILEPATH));
2664 (void)write(fd, &magic, sizeof(magic));
2665 (void)write(fd, &now, sizeof(now));
2666 (void)write(fd, &cbstuff, sizeof(cbstuff));
2667 (void)write(fd, TimeOuts, sizeof(TimeOuts));
2668 (void)write(fd, timeout, sizeof(timeout));
2669 (void)write(fd, &tfirst, sizeof(tfirst));
2670 freelisthead = cbtoi((struct CallBack *)CBfree);
2671 (void)write(fd, &freelisthead, sizeof(freelisthead)); /* This is a pointer */
2672 freelisthead = fetoi((struct FileEntry *)FEfree);
2673 (void)write(fd, &freelisthead, sizeof(freelisthead)); /* This is a pointer */
2674 (void)write(fd, HashTable, sizeof(HashTable));
2675 (void)write(fd, &CB[1], sizeof(CB[1]) * cbstuff.nblks); /* CB stuff */
2676 (void)write(fd, &FE[1], sizeof(FE[1]) * cbstuff.nblks); /* FE stuff */
2683 DumpCallBackState(void) {
2687 rc = DumpCallBackState_r();
2693 #endif /* !INTERPRET_DUMP */
2695 #ifdef INTERPRET_DUMP
2697 /* This is only compiled in for the callback analyzer program */
2698 /* Returns the time of the dump */
2700 ReadDump(char *file, int timebits)
2703 afs_uint32 magic, freelisthead;
2711 fd = open(file, oflag);
2713 fprintf(stderr, "Couldn't read dump file %s\n", file);
2716 read(fd, &magic, sizeof(magic));
2717 if (magic == MAGICV2) {
2720 if (magic != MAGIC) {
2722 "Magic number of %s is invalid. You might be trying to\n",
2725 "run this program on a machine type with a different byte ordering.\n");
2729 if (timebits == 64) {
2730 read(fd, &now64, sizeof(afs_int64));
2731 now = (afs_int32) now64;
2733 read(fd, &now, sizeof(afs_int32));
2735 read(fd, &cbstuff, sizeof(cbstuff));
2736 read(fd, TimeOuts, sizeof(TimeOuts));
2737 read(fd, timeout, sizeof(timeout));
2738 read(fd, &tfirst, sizeof(tfirst));
2739 read(fd, &freelisthead, sizeof(freelisthead));
2740 CB = ((struct CallBack
2741 *)(calloc(cbstuff.nblks, sizeof(struct CallBack)))) - 1;
2742 FE = ((struct FileEntry
2743 *)(calloc(cbstuff.nblks, sizeof(struct FileEntry)))) - 1;
2744 CBfree = (struct CallBack *)itocb(freelisthead);
2745 read(fd, &freelisthead, sizeof(freelisthead));
2746 FEfree = (struct FileEntry *)itofe(freelisthead);
2747 read(fd, HashTable, sizeof(HashTable));
2748 read(fd, &CB[1], sizeof(CB[1]) * cbstuff.nblks); /* CB stuff */
2749 read(fd, &FE[1], sizeof(FE[1]) * cbstuff.nblks); /* FE stuff */
2751 perror("Error reading dumpfile");
2758 #include "AFS_component_version_number.h"
2760 #include "AFS_component_version_number.c"
2763 static afs_uint32 *cbTrack;
2766 main(int argc, char **argv)
2768 int err = 0, cbi = 0, stats = 0, noptions = 0, all = 0, vol = 0, raw = 0;
2770 struct FileEntry *fe;
2771 struct CallBack *cb;
2775 memset(&fid, 0, sizeof(fid));
2778 while (argc && **argv == '-') {
2781 if (!strcmp(*argv, "-host")) {
2787 cbi = atoi(*++argv);
2788 } else if (!strcmp(*argv, "-fid")) {
2794 fid.Volume = atoi(*++argv);
2795 fid.Vnode = atoi(*++argv);
2796 fid.Unique = atoi(*++argv);
2797 } else if (!strcmp(*argv, "-time")) {
2798 fprintf(stderr, "-time not supported\n");
2800 } else if (!strcmp(*argv, "-stats")) {
2802 } else if (!strcmp(*argv, "-all")) {
2804 } else if (!strcmp(*argv, "-raw")) {
2806 } else if (!strcmp(*argv, "-timebits")) {
2812 timebits = atoi(*++argv);
2813 if ((timebits != 32)
2817 } else if (!strcmp(*argv, "-volume")) {
2823 vol = atoi(*++argv);
2828 if (err || argc != 1) {
2830 "Usage: cbd [-host cbid] [-fid volume vnode] [-stats] [-all] [-timebits 32"
2832 "] callbackdumpfile\n");
2834 "[cbid is shown for each host in the hosts.dump file]\n");
2837 now = ReadDump(*argv, timebits);
2838 if (stats || noptions == 0) {
2839 time_t uxtfirst = UXtime(tfirst), tnow = now;
2840 printf("The time of the dump was %u %s", (unsigned int) now, ctime(&tnow));
2841 printf("The last time cleanup ran was %u %s", (unsigned int) uxtfirst,
2843 PrintCallBackStats();
2846 cbTrack = calloc(cbstuff.nblks, sizeof(cbTrack[0]));
2851 struct CallBack *cb;
2852 struct FileEntry *fe;
2854 for (hash = 0; hash < FEHASH_SIZE; hash++) {
2855 for (feip = &HashTable[hash]; (fe = itofe(*feip));) {
2856 if (!vol || (fe->volid == vol)) {
2857 afs_uint32 fe_i = fetoi(fe);
2859 for (cb = itocb(fe->firstcb); cb; cb = itocb(cb->cnext)) {
2860 afs_uint32 cb_i = cbtoi(cb);
2862 if (cb_i > cbstuff.nblks) {
2863 printf("CB index out of range (%u > %d), stopped for this FE\n",
2864 cb_i, cbstuff.nblks);
2868 if (cbTrack[cb_i]) {
2869 printf("CB entry already claimed for FE[%u] (this is FE[%u]), stopped\n",
2870 cbTrack[cb_i], fe_i);
2873 cbTrack[cb_i] = fe_i;
2885 afs_uint32 cfirst = cbi;
2890 } while (cbi != cfirst);
2895 printf("No callback entries for %u.%u\n", fid.Volume, fid.Vnode);
2898 cb = itocb(fe->firstcb);
2901 cb = itocb(cb->cnext);
2906 for (i = 1; i < cbstuff.nblks; i++) {
2907 p = (afs_int32 *) & FE[i];
2908 printf("%d:%12x%12x%12x%12x\n", i, p[0], p[1], p[2], p[3]);
2917 PrintCB(struct CallBack *cb, afs_uint32 now)
2919 struct FileEntry *fe = itofe(cb->fhead);
2920 time_t expires = TIndexToTime(cb->thead);
2925 printf("vol=%" AFS_VOLID_FMT " vn=%u cbs=%d hi=%d st=%d fest=%d, exp in %lu secs at %s",
2926 afs_printable_VolumeId_lu(fe->volid), fe->vnode, fe->ncbs,
2927 cb->hhead, cb->status, fe->status, expires - now, ctime(&expires));
2932 #if !defined(INTERPRET_DUMP)
2934 ** try breaking calbacks on afidp from host. Use multi_rx.
2935 ** return 0 on success, non-zero on failure
2938 MultiBreakCallBackAlternateAddress(struct host *host, struct AFSCBFids *afidp)
2942 retVal = MultiBreakCallBackAlternateAddress_r(host, afidp);
2948 MultiBreakCallBackAlternateAddress_r(struct host *host,
2949 struct AFSCBFids *afidp)
2952 struct rx_connection **conns;
2953 struct rx_connection *connSuccess = 0;
2954 struct AddrPort *interfaces;
2955 static struct rx_securityClass *sc = 0;
2956 static struct AFSCBs tc = { 0, 0 };
2959 /* nothing more can be done */
2960 if (!host->interface)
2961 return 1; /* failure */
2963 /* the only address is the primary interface */
2964 if (host->interface->numberOfInterfaces <= 1)
2965 return 1; /* failure */
2967 /* initialise a security object only once */
2969 sc = rxnull_NewClientSecurityObject();
2971 i = host->interface->numberOfInterfaces;
2972 interfaces = calloc(i, sizeof(struct AddrPort));
2973 conns = calloc(i, sizeof(struct rx_connection *));
2974 if (!interfaces || !conns) {
2975 ViceLogThenPanic(0, ("Failed malloc in "
2976 "MultiBreakCallBackAlternateAddress_r\n"));
2979 /* initialize alternate rx connections */
2980 for (i = 0, j = 0; i < host->interface->numberOfInterfaces; i++) {
2981 /* this is the current primary address */
2982 if (host->host == host->interface->interface[i].addr &&
2983 host->port == host->interface->interface[i].port)
2986 interfaces[j] = host->interface->interface[i];
2988 rx_NewConnection(interfaces[j].addr,
2989 interfaces[j].port, 1, sc, 0);
2990 rx_SetConnDeadTime(conns[j], 2);
2991 rx_SetConnHardDeadTime(conns[j], AFS_HARDDEADTIME);
2995 opr_Assert(j); /* at least one alternate address */
2997 ("Starting multibreakcall back on all addr for host %p (%s:%d)\n",
2998 host, afs_inet_ntoa_r(host->host, hoststr), ntohs(host->port)));
3000 multi_Rx(conns, j) {
3001 multi_RXAFSCB_CallBack(afidp, &tc);
3005 if (host->callback_rxcon)
3006 rx_DestroyConnection(host->callback_rxcon);
3007 host->callback_rxcon = conns[multi_i];
3008 /* add then remove */
3009 addInterfaceAddr_r(host, interfaces[multi_i].addr,
3010 interfaces[multi_i].port);
3011 removeInterfaceAddr_r(host, host->host, host->port);
3012 host->host = interfaces[multi_i].addr;
3013 host->port = interfaces[multi_i].port;
3014 connSuccess = conns[multi_i];
3015 rx_SetConnDeadTime(host->callback_rxcon, 50);
3016 rx_SetConnHardDeadTime(host->callback_rxcon, AFS_HARDDEADTIME);
3018 ("multibreakcall success with addr %s:%d\n",
3019 afs_inet_ntoa_r(interfaces[multi_i].addr, hoststr),
3020 ntohs(interfaces[multi_i].port)));
3027 /* Destroy all connections except the one on which we succeeded */
3028 for (i = 0; i < j; i++)
3029 if (conns[i] != connSuccess)
3030 rx_DestroyConnection(conns[i]);
3036 return 0; /* success */
3038 return 1; /* failure */
3043 ** try multi_RX probes to host.
3044 ** return 0 on success, non-0 on failure
3047 MultiProbeAlternateAddress_r(struct host *host)
3050 struct rx_connection **conns;
3051 struct rx_connection *connSuccess = 0;
3052 struct AddrPort *interfaces;
3053 static struct rx_securityClass *sc = 0;
3056 /* nothing more can be done */
3057 if (!host->interface)
3058 return 1; /* failure */
3060 /* the only address is the primary interface */
3061 if (host->interface->numberOfInterfaces <= 1)
3062 return 1; /* failure */
3064 /* initialise a security object only once */
3066 sc = rxnull_NewClientSecurityObject();
3068 i = host->interface->numberOfInterfaces;
3069 interfaces = calloc(i, sizeof(struct AddrPort));
3070 conns = calloc(i, sizeof(struct rx_connection *));
3071 if (!interfaces || !conns) {
3072 ViceLogThenPanic(0, ("Failed malloc in "
3073 "MultiProbeAlternateAddress_r\n"));
3076 /* initialize alternate rx connections */
3077 for (i = 0, j = 0; i < host->interface->numberOfInterfaces; i++) {
3078 /* this is the current primary address */
3079 if (host->host == host->interface->interface[i].addr &&
3080 host->port == host->interface->interface[i].port)
3083 interfaces[j] = host->interface->interface[i];
3085 rx_NewConnection(interfaces[j].addr,
3086 interfaces[j].port, 1, sc, 0);
3087 rx_SetConnDeadTime(conns[j], 2);
3088 rx_SetConnHardDeadTime(conns[j], AFS_HARDDEADTIME);
3092 opr_Assert(j); /* at least one alternate address */
3094 ("Starting multiprobe on all addr for host %p (%s:%d)\n",
3095 host, afs_inet_ntoa_r(host->host, hoststr),
3096 ntohs(host->port)));
3098 multi_Rx(conns, j) {
3099 multi_RXAFSCB_ProbeUuid(&host->interface->uuid);
3103 if (host->callback_rxcon)
3104 rx_DestroyConnection(host->callback_rxcon);
3105 host->callback_rxcon = conns[multi_i];
3106 /* add then remove */
3107 addInterfaceAddr_r(host, interfaces[multi_i].addr,
3108 interfaces[multi_i].port);
3109 removeInterfaceAddr_r(host, host->host, host->port);
3110 host->host = interfaces[multi_i].addr;
3111 host->port = interfaces[multi_i].port;
3112 connSuccess = conns[multi_i];
3113 rx_SetConnDeadTime(host->callback_rxcon, 50);
3114 rx_SetConnHardDeadTime(host->callback_rxcon, AFS_HARDDEADTIME);
3116 ("multiprobe success with addr %s:%d\n",
3117 afs_inet_ntoa_r(interfaces[multi_i].addr, hoststr),
3118 ntohs(interfaces[multi_i].port)));
3123 ("multiprobe failure with addr %s:%d\n",
3124 afs_inet_ntoa_r(interfaces[multi_i].addr, hoststr),
3125 ntohs(interfaces[multi_i].port)));
3127 /* This is less than desirable but its the best we can do.
3128 * The AFS Cache Manager will return either 0 for a Uuid
3129 * match and a 1 for a non-match. If the error is 1 we
3130 * therefore know that our mapping of IP address to Uuid
3131 * is wrong. We should attempt to find the correct
3132 * Uuid and fix the host tables.
3134 if (multi_error == 1) {
3135 /* remove the current alternate address from this host */
3137 removeInterfaceAddr_r(host, interfaces[multi_i].addr, interfaces[multi_i].port);
3141 #ifdef AFS_DEMAND_ATTACH_FS
3142 /* try to bail ASAP if the fileserver is shutting down */
3144 if (fs_state.mode == FS_MODE_SHUTDOWN) {
3153 /* Destroy all connections except the one on which we succeeded */
3154 for (i = 0; i < j; i++)
3155 if (conns[i] != connSuccess)
3156 rx_DestroyConnection(conns[i]);
3162 return 0; /* success */
3164 return 1; /* failure */
3167 #endif /* !defined(INTERPRET_DUMP) */